Efficient Design Flow for Fixed-Point Systems

Abstract

The complexity of DSP systems grows at an ever-increasing rate while the implementation of these designs must meet criteria like minimum cost and a short time to market. Hence there is a growing need for efficient design automation and a seamless design flow that allows the execution of the design steps at the highest suitable level of abstraction.Especially for the design of fixed-point algorithms and systems, tool support is very rare, though for most digital systems the design has to result in a fixed-point implementation. This is due to the fact that these systems are sensitive to power consumption, chip size and price per device. Fixed-point realizations outperform floating-point realizations by far with regard to these criteria. On the other hand, algorithm design starts from a floating-pointdescription in order to initially abstract from all fixed-point effects. The resulting gap between the floating-point prototype and the fixed-point implementation represents one of the major bottlenecks in today’s digital designs.This paper will give a survey of FRIDGE1, a tool suite that permits a seamless design flow, starting from an ANSI-C floating-point algorithm which is then converted to a fixed-point description in fixed-C2. This bit-true description of the algorithm can finally be mapped to different implementation targets, programmable DSPs or dedicated hardware structures.KeywordsDesign FlowHost MachineLocal AnnotationGlobal AnnotationComplex Design SpaceThese keywords were added by machine and not by the authors. This process is experimental and the keywords may be updated as the learning algorithm improves.